These numbers always stun me, no matter how often I see them. Full report here. Thumbnail of international comparisons below. The state of Maine, at 273 prisoners per 100,000 residents has the lowest incarceration rate of any American state, and comes in between Moldova and Latvia in the international comparisons.

Australia comes in at 130/100,000, about the same as the Netherlands.

Oaksterdam University.

A blizzard just whacked the East Coast of the US, and we’re bracing for a winter storm here in Northern California. And so you open up the New York Times to see…

The Times has a nice slide show here too.

So there I am, cruising the Sydney Morning Herald website, and the banner ad is from the Republican National Committee. That could well be a first: an Australian sourced web site piping me ads from an American source.

From HuffPost, an article on how the longer the primary battle goes on, the harder it will be for the Democrats to fundraise for the general election campaign. In the middle of the piece though is this:

..the DNC, nevertheless, ended 2007 nearly broke, after embarking on two expensive projects — the building of a national voter file and a 50-state staffing program — according to reports filed at the Federal Election Commission.

…the DNC has developed and refined a national voter list at a cost of more than $10 million.

That strikes me as a staggering amount of money, and I suspect they are up to more than a just a voter list. You can probably make a decent start here. Its the matching with useful/interesting consumer data that must be making this so expensive, and then maintaining the lists.

In Australia, this list is supplied to the parties on request by the Australian Electoral Commission. Merging in some data from vendors like Acxiom might run up the bill a little, but nothing like $10M. Just another one of the perks of having a system of compulsory voting.

Turnbull tries to get Swan to step in a pile of dog poo re NAIRU. It doesn’t really matter in a sense, since The Sydney Morning Herald can’t get it right, even in a joke about the obscurity of the term; see the caption on the photo-illustration below, from their web site.

The layoffs at Fairfax are starting to show: Gittens must have been busy when they pushed this to the web.

The Apology to the Stolen Generations of Australian Aborigines garnered space on the front page of the New York Times yesterday. Click on the thumbnail.

For the record, I thought Rudd’s speech was amazing, one of the more emotionally resonant things I’ve ever heard come out of the mouth of an Australian politician. A little preachy for my taste (and in this respect contrast Keating’s Redfern oration), but its tough for The Big Speech on The Apology not to be preachy.

It was also terrific political theatre. Under a call to bipartisanship, Rudd and Labor knee-capped the Opposition. I mean, what a way to start the parliamentary term of the new government: a morally-charged, emotionally resonant speech that appears to have “stopped the nation”, standing ovations in the parliament, you “right the wrongs” of the last decade, an apology to the Stolen Generations at 9am, and sounding the end of Workchoices at 11.30am.

And so poor old Brendan Nelson was really in a bind. He is the newly installed leader of a party that is at best conflicted on issues to do with reconciliation. Under those circumstances, I thought the speech he gave was pretty good, and that he carried himself with a fair amount of grace through the whole thing. He could really be nothing other than Rudd’s dance partner; the worse thing he could have done was step on Rudd’s feet, which he didn’t.

From Nicholas Kristof in the NYTimes:

Mr. Obama does surprisingly well among evangelical Christians, an important constituency in swing states. For example, Relevant magazine, which caters to young evangelicals, asked its readers: “Who would Jesus vote for?” Mr. Obama was the winner and came out 27 percentage points ahead of Mrs. Clinton.

I think they’ve got it wrong. Jesus isn’t a voter.  Jesus is a Superdelegate…

And it is really awesome that those readers of Relevant actually know how He would vote.

We’re a step closer. Safari 3.1 will support it (inter alia), and see here for some background.

But how will it render under IE?

A fellow political methodologist asks:

I have been asked to do some forecasting… I have a vague idea of how to do this in WinBUGS, but was wondering if you have done such a model (I’m assuming you have given your interests). If so, do you have an sample-esque code you could send along.

Briefly, I’m predicting the probability of a certain type of event occuring given a set of covariates. The data are dichotomous with a lagged DV and some co-variates. It’s a trivial model to estimate, but I’m not exactly sure what to add to ‘forecast’ events.

I’m guessing it would be something like:

p_new <- mu[new]
mu[new] <- ???

The BUGS way of doing things makes this kind of easy. Suppose you've got data on points, 1, ..., R, and you want forecasts for R+1, ..., T. Then you can just write a loop for 1:T, and the fact that y[R+1:T] are unobserved nodes is really not problematic for the software, provided all the parents of those unobserved nodes are specified. Take a trivial case, a Gaussian random walk with unit innovation variance, and standard Gaussian initializing the walk:

for(i in 2:T){
  y[i] ~ dnorm(y[i-1],1)
}

y[1] ~ dnorm(0,1)  ## initial conditions

Now as far as BUGS is concerned, it doesn't matter if all or any of the y[i] are observed. The initialization for y[1] kicks things off, and away you go.

Generalize/modify this as appropriate for (a) covariates in the model for y; (b) discrete y. For the case with covariates X contemporaneous with y, you'll want some kind of prior or model to come up with values of X[R+1:T], since these are parents of y[R+1:T] and the program won't know what to do without some kind of model/prior for them.

The other thing here is that when you feed data to BUGS, you'll supply data for y[1:T], but with NAs as the "data" for y[R+1:T]. Similarly, pad out X with NAs for missing or yet-to-be-realized observations, but again, you'll need a model and/or prior for those elements of X (again, since they are parents of y). Finally, for the "missing" y and X, you can supply initial values: if you are going to do this, you create initial values for the entire y vector and X matrix, but with NAs as the initial values for elements of those objects that are actually observed.